Embedding API

Protocol protocols published

Also known as: Embeddings Endpoint, Text Embedding API

Definition

An API endpoint that converts text into vector embeddings—dense numerical representations suitable for similarity search, clustering, and classification. Embedding APIs take text strings and return arrays of floats representing the text in a high-dimensional semantic space.

What this is NOT

  • Not the embedding model (API is the interface)
  • Not chat completions (different purpose)
  • Not the vector database (that stores embeddings)

Alternative Interpretations

Different communities use this term differently:

llm-practitioners

The /v1/embeddings endpoint (OpenAI) or equivalent that takes text and returns embedding vectors. Used for RAG, semantic search, and other embedding-based applications.

Sources: OpenAI Embeddings documentation, Cohere Embed API, Voyage AI documentation

Examples

  • POST /v1/embeddings with text input
  • OpenAI text-embedding-3-small returning 1536-dim vectors
  • Cohere embed-v3 with input_type parameter
  • Batch embedding of 100 documents

Counterexamples

Things that might seem like Embedding API but are not:

  • Chat completions API (generates text, not embeddings)
  • The embedding model itself (API is the interface)
  • Vector database query (uses embeddings, doesn't create them)

Relations

  • specializes openai-api (Embeddings is part of the OpenAI API family)
  • produces embedding (API produces embedding vectors)
  • requires vector-search (Embeddings enable vector search)

See Also